Uncertainty quantification of extreme precipitation projections including regional climate model interdependency and non-stationary bias

Abstract:

Extreme precipitation projections estimated from regional climate models are subject to large uncertainties. Several probabilistic procedures based on multi-model ensembles have been suggested in the literature to quantify the uncertainty. These procedures often require several assumptions. Two common assumptions are that the climate models are independent and that changes in climate model biases are negligible. However, there is evidence suggesting that these assumptions might not be valid. This study presents a Bayesian framework that accounts for model dependencies and changes in model biases. The Bayesian framework is used to estimate the uncertainty in extreme precipitation projections over Denmark with and without making these assumptions based on the multi-model ensemble of regional climate models from the ENSEMBLES project.

The evaluation of the two assumptions confirms that the regional climate models cannot be considered independent and shows that the bias depends on the precipitation amount. The results from the Bayesian framework show that this has a large influence on the uncertainty estimated. If the interdependency and change in bias are not taken into account, the results may be treated as more precise than they really are and underestimate the climate change impact on extreme precipitation. This study highlights the importance of investigating the underlying assumptions in climate change impact studies and suggests a Bayesian framework which is flexible and allows testing different assumptions and their influence.